C Crucible contact sheet
design system ↗ palette notes ↗

Crucible · 16 slices

tier 1 spine (6) · tier 2 blue+measure (4) · tier 3 ops + exports (6) · click any tile to open
Tier 1 · spine of the demo
01
launcher
Run Launcher
spec pickers · oracle mix · cost ceiling · launch a new red↔blue round
/runs/new
02
central
Live Run View
streams oracle verdicts · ASR · costs · traces — the screen DS chrome is built around
/runs/:runId
04
headline · outcomes
Outcomes
undetected-hack · v↔h gap · recall · cost/hack · human-min/1k · ASR + detection curves
/dashboard
03
drill-down
Verdict Detail
per-oracle: held-out · metamorphic · differential · fuzz · judge — built on AuditTraceCard
/verdicts/:vid
05
audit
Audit Row Replayer
audit trace · check fired · obligation · 1-click replay
/audit/:rowId
06
memory
Strategy Catalog
persistent memory of winning tactics across runs · sortable, taggable
/strategies
Tier 2 · blue + measure
07
blue loop
Blue Patch Review
apply-patch flow · retrain · re-eval on held-out · 4-eyes acks
/patches/:pid
09
measure
Co-evolution Curves
red↔blue over N rounds · convergence · oscillations · ΔASR per patch
/dashboard/co-evolution
08
halt
Halt Certification
runs blocked · lift conditions · 4-eyes override · halt history
/halts/:hid
10
0.92recall
measure
White-box Self-test
scheme revealed not instances · recall as a measured number, not a claim
/whitebox/:wbId
Tier 3 · operations & exports
11
ops
Health
9 leaves · overall degraded strip · open incidents
/health
12
ops
Admin · Debug
mock-llm toggle · producer routing · halt bypass · feature flags
/admin/debug
13
export
Benchmark · Leaderboard
seeded corpus · reproduce CLI · 14 submissions · overfit-flag
/benchmarks/seeded-v3
14
export · print
SR 11-7 Model Risk Report
bank-audience export · printable · dual sign-off
/reports/sr-11-7
15
governance
Workspace · Roles & Policy
5 roles · 12 members · 14 active policies · cost-ledger drawer
/workspaces/acme-fraud
16
governance
Sealed-spec History
timeline · diff view · signatures · provenance chain
/specs
handoff status
16 of 16 tier-items covered · wordmark on every slice links back here · breadcrumbs in 09–16 wired to dashboard / canvas · slice-14 is print-styled (Cmd-P → PDF).
known stubs
deep-link hrefs inside slices (e.g. r_8f3a, p_2a17, h_19c4) are still # — wire when the route shape settles. Empty / loading states not drawn. Mobile out of scope.